vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : increased max_nodes as large MoE models use massive amounts of nodes during warmup

Browse source
This commit is contained in:
Stanisław Szymczyk 2025-02-01 12:43:14 +01:00
parent 83a473a001
commit c8bc6e4ff4
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
src/llama-model.cpp
View file

@ -3552,7 +3552,7 @@ size_t llama_model::size() const {
}
size_t llama_model::max_nodes() const {
return std::max<size_t>(8192, tensors_by_name.size()*5);
return std::max<size_t>(65536, tensors_by_name.size()*5);
}
size_t llama_model::n_devices() const {